2023-11-14 13:35:43 +00:00
|
|
|
#include "ttm.h"
|
|
|
|
|
|
|
|
void ttm(
|
|
|
|
const int transB, const int mode,
|
|
|
|
const int* dimA, const int ordA, const int nrowB, const int ncolB,
|
|
|
|
const double alpha,
|
|
|
|
const double* A,
|
|
|
|
const double* B, const int ldB, // TODO: ldB is IGNORED!!!
|
|
|
|
const double beta,
|
|
|
|
double* C
|
|
|
|
) {
|
|
|
|
|
|
|
|
// Strides are the "leading" and "trailing" dimensions of the matricized
|
|
|
|
// tensor `A` in the following matrix-matrix multiplications
|
|
|
|
// `stride[0] <- prod(dim(A)[seq_len(mode - 1)])`
|
|
|
|
// `stride[1] <- dim(A)[mode]`
|
|
|
|
// `stride[2] <- prod(dim(A)[-seq_len(mode)])`
|
|
|
|
int stride[3] = {1, dimA[mode], 1};
|
|
|
|
for (int i = 0; i < ordA; ++i) {
|
|
|
|
stride[0] *= (i < mode) ? dimA[i] : 1;
|
|
|
|
stride[2] *= (i > mode) ? dimA[i] : 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (mode == 0) {
|
|
|
|
// mode 1: C = alpha (A x_1 op(B))_(1) + beta C
|
|
|
|
// = alpha op(B) A_(1) + beta C
|
|
|
|
// as a single Matrix-Matrix multiplication
|
|
|
|
F77_CALL(dgemm)(transB ? "T" : "N", "N",
|
|
|
|
(transB ? &ncolB : &nrowB), &stride[2], &stride[1], &alpha,
|
|
|
|
B, &nrowB, A, &stride[1],
|
|
|
|
&beta, C, (transB ? &ncolB : &nrowB)
|
|
|
|
FCONE FCONE);
|
|
|
|
} else {
|
|
|
|
// Other modes can be written as blocks of matrix multiplications
|
|
|
|
// C_:,:,i2 = alpha (A x_m op(B))_(m)' + beta C_:,:,i2
|
|
|
|
// = alpha A_(m)' op(B)' + beta C_:,:,i2
|
|
|
|
for (int i2 = 0; i2 < stride[2]; ++i2) {
|
|
|
|
F77_CALL(dgemm)("N", transB ? "N" : "T",
|
|
|
|
&stride[0], (transB ? &ncolB : &nrowB), &stride[1], &alpha,
|
|
|
|
&A[i2 * stride[0] * stride[1]], &stride[0], B, &nrowB,
|
|
|
|
&beta, &C[i2 * stride[0] * (transB ? ncolB : nrowB)], &stride[0]
|
|
|
|
FCONE FCONE);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
// (reference implementation)
|
|
|
|
// Tensor Times Matrix / Mode Product for `op(B) == B`
|
|
|
|
memset(c, 0, sizeC * sizeof(double));
|
|
|
|
for (int i2 = 0; i2 < stride[2]; ++i2) {
|
|
|
|
for (int i1 = 0; i1 < stride[1]; ++i1) { // stride[1] == ncols(B)
|
|
|
|
for (int j = 0; j < nrow; ++j) {
|
|
|
|
for (int i0 = 0; i0 < stride[0]; ++i0) {
|
|
|
|
c[i0 + (j + i2 * nrow) * stride[0]] +=
|
|
|
|
a[i0 + (i1 + i2 * stride[1]) * stride[0]] * b[j + i1 * nrow];
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
*/
|
|
|
|
}
|
|
|
|
|
2022-04-29 14:50:51 +00:00
|
|
|
|
|
|
|
/**
|
|
|
|
* Tensor Times Matrix a.k.a. Mode Product
|
|
|
|
*
|
2022-05-06 20:28:08 +00:00
|
|
|
* @param A multi-dimensional array
|
2022-04-29 14:50:51 +00:00
|
|
|
* @param B matrix
|
|
|
|
* @param m mode index (1-indexed)
|
2022-10-06 12:25:40 +00:00
|
|
|
* @param op boolean if `B` is transposed
|
2022-04-29 14:50:51 +00:00
|
|
|
*/
|
2023-11-14 13:35:43 +00:00
|
|
|
extern SEXP R_ttm(SEXP A, SEXP B, SEXP m, SEXP op) {
|
2022-04-29 14:50:51 +00:00
|
|
|
|
|
|
|
// get zero indexed mode
|
2022-10-06 12:25:40 +00:00
|
|
|
const int mode = Rf_asInteger(m) - 1;
|
2022-04-29 14:50:51 +00:00
|
|
|
|
|
|
|
// get dimension attribute of A
|
2023-11-14 13:35:43 +00:00
|
|
|
SEXP dimA = Rf_getAttrib(A, R_DimSymbol);
|
2022-10-06 12:25:40 +00:00
|
|
|
|
|
|
|
// operation on `B` (transposed or not)
|
2023-11-14 13:35:43 +00:00
|
|
|
const int transB = Rf_asLogical(op);
|
2022-10-06 12:25:40 +00:00
|
|
|
|
|
|
|
// as well as `B`s dimensions
|
2023-11-14 13:35:43 +00:00
|
|
|
const int nrowB = Rf_nrows(B);
|
|
|
|
const int ncolB = Rf_ncols(B);
|
2022-04-29 14:50:51 +00:00
|
|
|
|
|
|
|
// validate mode (mode must be smaller than the nr of dimensions)
|
2023-11-14 13:35:43 +00:00
|
|
|
if (mode < 0 || Rf_length(dimA) <= mode) {
|
2022-10-06 12:25:40 +00:00
|
|
|
Rf_error("Illegal mode");
|
2022-04-29 14:50:51 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
// and check if B is a matrix of non degenetate size
|
2022-10-06 12:25:40 +00:00
|
|
|
if (!Rf_isMatrix(B)) {
|
|
|
|
Rf_error("Expected a matrix as second argument");
|
2022-04-29 14:50:51 +00:00
|
|
|
}
|
2022-10-06 12:25:40 +00:00
|
|
|
if (!Rf_nrows(B) || !Rf_ncols(B)) {
|
|
|
|
Rf_error("Zero dimension detected");
|
2022-04-29 14:50:51 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
// check matching of dimensions
|
2023-11-14 13:35:43 +00:00
|
|
|
if (INTEGER(dimA)[mode] != (transB ? nrowB : ncolB)) {
|
2022-10-06 12:25:40 +00:00
|
|
|
Rf_error("Dimension missmatch");
|
2022-04-29 14:50:51 +00:00
|
|
|
}
|
|
|
|
|
2023-11-14 13:35:43 +00:00
|
|
|
// calc nr of response elements (size of C)
|
|
|
|
// `prod(dim(C)) = prod(dim(A)[-mode]) * nrow(if(transB) t(B) else B)`
|
2022-05-06 20:28:08 +00:00
|
|
|
int sizeC = 1;
|
2023-11-14 13:35:43 +00:00
|
|
|
for (int i = 0; i < Rf_length(dimA); ++i) {
|
|
|
|
int size = INTEGER(dimA)[i];
|
2022-04-29 14:50:51 +00:00
|
|
|
// check for non-degenetate dimensions
|
|
|
|
if (!size) {
|
2022-10-06 12:25:40 +00:00
|
|
|
Rf_error("Zero dimension detected");
|
2022-04-29 14:50:51 +00:00
|
|
|
}
|
2023-11-14 13:35:43 +00:00
|
|
|
sizeC *= (i == mode) ? (transB ? ncolB : nrowB) : size;
|
2022-04-29 14:50:51 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
// create response object C
|
2022-10-06 12:25:40 +00:00
|
|
|
SEXP C = PROTECT(Rf_allocVector(REALSXP, sizeC));
|
2022-04-29 14:50:51 +00:00
|
|
|
|
|
|
|
// Tensor Times Matrix / Mode Product
|
2023-11-14 13:35:43 +00:00
|
|
|
ttm(transB, mode,
|
|
|
|
INTEGER(dimA), Rf_length(dimA), nrowB, ncolB,
|
|
|
|
1.0, REAL(A), REAL(B), nrowB, 0.0, REAL(C));
|
2022-04-29 14:50:51 +00:00
|
|
|
|
|
|
|
// finally, set result dimensions
|
2023-11-14 13:35:43 +00:00
|
|
|
SEXP dimC = PROTECT(Rf_allocVector(INTSXP, Rf_length(dimA)));
|
|
|
|
for (int i = 0; i < Rf_length(dimA); ++i) {
|
|
|
|
INTEGER(dimC)[i] = (i == mode) ? (transB ? ncolB : nrowB) : INTEGER(dimA)[i];
|
2022-04-29 14:50:51 +00:00
|
|
|
}
|
2023-11-14 13:35:43 +00:00
|
|
|
Rf_setAttrib(C, R_DimSymbol, dimC);
|
2022-04-29 14:50:51 +00:00
|
|
|
|
|
|
|
// release C to the hands of the garbage collector
|
|
|
|
UNPROTECT(2);
|
|
|
|
|
|
|
|
return C;
|
|
|
|
}
|